High-throughput intracellular biopsy of microRNAs for dissecting the temporal dynamics of cellular heterogeneity

See allHide authors and affiliations

Science Advances  10 Jun 2020:
Vol. 6, no. 24, eaba4971
DOI: 10.1126/sciadv.aba4971


The capability to analyze small RNAs responsible for post-transcriptional regulation of genes expression is essential for characterizing cellular phenotypes. Here, we describe an intracellular biopsy technique (inCell-Biopsy) for fast, multiplexed, and highly sensitive profiling of microRNAs (miRNAs). The technique uses an array of diamond nanoneedles that are functionalized with size-dependent RNA binding proteins, working as “fishing rods” to directly pull miRNAs out of cytoplasm while keeping the cells alive, thus enabling quasi-single-cell miRNA analysis. Each nanoneedle works as a reaction chamber for parallel in situ amplification, visualization, and quantification of miRNAs as low as femtomolar, which is sufficient to detect miRNAs of a single-copy intracellular abundance with specificity to single-nucleotide variation. Using inCell-Biopsy, we analyze the temporal miRNA transcriptome over the differentiation of embryonic stem cells (ESCs). The combinatorial miRNA expression patterns derived by inCell-Biopsy identify emerging cell subpopulations differentiated from ESCs and reveal the dynamic evolution of cellular heterogeneity.


Embryonic stem cells (ESCs) are self-renewable and can differentiate into all types of cells in an adult organism. They are increasingly used in disease modeling, drug discovery, and regenerative medicine (13). More recently, the emergence of induced pluripotent stem cells (iPSCs) eludes the potential of patient-specific regeneration of damaged or diseased tissues (4, 5) and brings stem cell–based therapy to the forefront of the development of previously unidentified treatments for many diseases that are challenging for traditional methods. One of the issues for these efforts is the development of protocols to ensure directed differentiation of stem cells both in vitro and in vivo (6), which relies on the understanding and controlling of the cellular heterogeneity generated from stem cell differentiation and is critical for the clinical adoption of relevant therapeutic strategies (7).

MicroRNAs (miRNAs), a class of noncoding small RNAs, were reported to be involved in the regulation of self-renewal and differentiation of many stem cells (8). They regulate gene expression by binding to specific mRNA targets to promote mRNA degradation or translational inhibition. Many recent studies have shown a close relationship between miRNA transcriptome and cellular heterogeneity in different tissues (9, 10) or over stem cell differentiation (11, 12). However, in situ profiling of miRNAs in living cells is still challenging, hindering the adoption of miRNA as a therapeutic indicator in clinical practices (13). The trace amount of miRNAs in cytoplasm requires a profiling technique to be highly specific and sensitive (13, 14). A size selection step is typically required, following the isolation of total RNA from cellular homogenate (13, 15). Because of the short length of miRNAs, small primers are required in polymerase chain reaction (PCR) and often cause reduced priming efficiency, nonspecific hybridization, and, thus, erroneous results (14).

Among the tools for miRNA profiling, quantitative reverse transcription PCR (qRT-PCR) is the “gold standard,” but it is rather a validation instead of discovery tool. Deep sequencing resolves miRNA transcriptome with an extremely high throughput, but it is, however, disadvantaged by high cost, long turnover time, and complex data analysis (13). Both methods inherit similar issues from the required PCR process as mentioned above. Alternatively, microarray provides multiplexed miRNA analysis based on probe-target hybridization but has lower specificity and sensitivity (13, 14). All these commonly used techniques require isolation of RNA sample from cellular lysates, providing only average measurements of miRNAs for all cells. Consequently, important information about the heterogeneity of the cell populations would be missing and is only accessible by using single cell–based analytical techniques (16). Some in situ miRNA assay tools were recently reported by combining high-resolution microscopy with nanotechnology (17, 18), which are less quantitative and sometimes limited by toxicity issues.

Here, we describe an intracellular biopsy (inCell-Biopsy) technique for multiplexed in situ profiling of miRNAs in living cells. An array of diamond nanoneedles were functionalized with RNA binding protein (p19) and were used as “fishing rods” to directly pull multiple targeted miRNAs out of cell cytoplasm in a few minutes, leaving the cells alive. After the inCell-Biopsy operation, each nanoneedle then worked as a separated reaction plant for parallel in situ amplification, visualization, and quantification of miRNAs. The detection limit can reach as low as 10−15 M, which is almost three orders of magnitude lower than the abundance of a single copy in a cell. Using inCell-Biopsy, we demonstrated multiplexed profiling of miRNAs in living cells and analyzed the temporal miRNA transcriptome over the differentiation of ESCs toward motor neurons, revealing the cellular heterogeneity and associated evolutionary dynamics of the differentiated cell populations based on miRNA expression.


Intracellular biopsy of miRNAs

The inCell-Biopsy technique is based on the continuous development of a “molecular fishing” system (19), which uses an array of diamond nanoneedles as fishing rods for minimum-invasive and reversible access of cytoplasmic regions of mammalian cells (Fig. 1A and movie S1). Specifically, for “fishing” miRNAs, a size-dependent RNA binding protein, p19, was cross-linked to functionalize the nanoneedles, working as the “fishing hook” to capture double-strand RNAs (dsRNAs) (Fig. 1A; more details in fig. S1). P19 can selectively bind to all dsRNAs of 20 to 22 base pairs (bp) (20, 21), which is a range covering almost all miRNAs in mammalian cells. As mature intracellular miRNAs are mostly single stranded (22), in this study, for each targeted miRNA, a “bait” RNA sequence was delivered to the cytoplasm to hybridize with the targets for p19 to capture. When the functionalized diamond nanoneedles are interfaced with live cells using a centrifugation facilitated procedure, the fishing rods can penetrate the cell membrane to access the cytoplasmic region via a temporary membrane disruption, which simultaneously facilitates the intracellular delivery of the bait RNAs (fig. S1) (19, 23). Upon the retrieval of the nanoneedles, the targeted miRNAs are isolated, leaving the cells alive (viability, 96.2 ± 1.5%; means ± SD, n = 3).

Fig. 1 Design of the inCell-Biopsy technique for miRNA profiling.

(A) Schematic illustration of intracellular biopsy of miRNAs from live cells. (B) On-needle amplification of miRNA signals by hybridization chain reaction (HCR). (C) DNase-assisted multiple rounds of signal visualization. (D) Image processing and informatic approaches for miRNA transcriptome analysis.

After the inCell-Biopsy operation, a hybridization chain reaction (HCR) was performed on the nanoneedles to amplify the miRNAs. The HCR is featured by two single-strand DNAs (ssDNAs) with a stem-loop structure (hairpin 1 and 2), which can cyclically hybridize with each other if the stem part of one sequence is open (Fig. 1B) (24). For each miRNA target, the corresponding bait RNA contains the complementary sequence plus a small encoding overhang part at its 3′ end, which binds to a small DNA sequence (initiator) to trigger the HCR (Fig. 1B). To enhance the detection sensitivity, one of the hairpins (hairpin 1) was fluorescently labeled and was quenched until its stem opening in the HCR. For multiplexing, the overhang part of the bait sequence was uniquely encoded for different miRNA targets, so that multiple miRNA targets can be visualized by different fluorophores (table S1). While the number of optically separable fluorophores may be limited (e.g., four channels), we implemented multiple rounds of HCR by removing the DNA hairpins after collecting the signals at the end of each round using deoxyribonuclease I (DNase I) enzyme, so that a new set of miRNAs could be examined to improve the analytical throughput of the inCell-Biopsy technique (Fig. 1C). Three rounds of HCR enabled us to examine 12 miRNAs after each biopsy. To enhance assay reliability, we dedicated one of the four channels to a reference miRNA (Fig. 1D), cel-miR-39, which does not exist in human or rodent cells and was artificially introduced to the cell cytoplasm, thus to eliminate potential systemic errors caused by experimental variations. The expression level of a targeted miRNA was indicated by the normalized fluorescence (with respect to the reference) acquired by confocal microscopy. While the nanoneedles could not be correlated to each cell with one-to-one mapping, the scattered signal from thousands of nanoneedles still retains the rich information about the cell population based on their miRNA expression (Fig. 1D).

Characterization of miRNA detection limit and specificity

To characterize the detection limit of our inCell-Biopsy technique, we first performed a mock experiment by profiling miRNAs from medium containing a premixed dsRNA (miR-34a and corresponding bait sequence) at different concentrations varying from 10−16 to 10−10 M. A nanoneedle chip was incubated in the solution and then rinsed before proceeding to further analysis. For every chip, signals from more than 1000 nanoneedles were collected and quantified (Fig. 2, A and B). Our results showed reliable differentiation of dsRNA of tested concentrations down to 10−15 M (Fig. 2C). The overall profile of the fluorescence intensities was observed to positively associate with the miRNA concentrations (Fig. 2D), and the ratio of positive nanoneedles also exhibited an association with the miRNA concentration, roughly following the Langmuir isotherm model (Fig. 2E; also see note S1) (25). To test the detection specificity, we performed an assay to detect the synthetic let-7a miRNA over sequences with 1–nucleotide (nt) mismatch (let-7c) or 2-nt mismatches (let-7b), and showed that the inCell-Biopsy technique is specific to single-nucleotide variation by successfully discriminating closely related miRNA sequences (Fig. 2F). The inCell-Biopsy was then implemented to detect miRNAs (let-7a or miR-34a) in cultured A549 cells (Fig. 2G). The two miRNAs were reported to express with different abundance in the cells: Let-7a is highly expressed, and miR-34a is of relatively low intracellular level (26). The nanoneedle chip was interfaced with A549 cells using a centrifugation-controlled method to initiate the intracellular biopsy (19, 23) and to deliver the two bait sequences to the intracellular domain. After a 15-min fishing reaction, the chip was retrieved from cells for analysis. For successful miRNA biopsy, the nanoneedles were identified to colocalize with the fluorescence signals from HCR amplification (Fig. 2H), which is significantly higher than different controls (Fig. 2, I and J; also see fig. S2A). The HCR amplification is especially useful in the detection of miRNAs of low intracellular abundance (e.g., miR-34a), which would otherwise be unobservable without the on-needle HCR amplification (fig. S2B).

Fig. 2 Technical characterization of inCell-biopsy.

(A) SEM image of the diamond nanoneedles; scale bar, 50 μm. (B) Fluorescence images (top view) showing miRNA signals (red) on the nanoneedles (green); scale bar, 50 μm. For (A) and (B), the boxed region is enlarged below; scale bars, 10 μm. (C) Analysis of detection limit with violin plots showing the distribution of miRNA signals from all nanoneedles. *P < 0.005 by Kruskal-Wallis test. a.u., arbitrary units. (D) Relationship between miRNA concentration and fluorescence averaged from all nanoneedles. The red line indicates logarithmic fit (R2 = 0.99, P < 0.001 by F test). (E) Relationship between miRNA concentration and ratio of signal+ nanoneedles. The blue dashed line indicates a nonlinear fit by Langmuir isotherm model (R2 = 0.98, P < 0.001 in F test). (F) Analysis of detection specificity. NC, no target included. *P < 0.001 by ANOVA test. (G) Image of A549 cells after treatment; scale bar, 50 μm. (H) Fluorescence visualization of miRNAs (red, let-7a or miR-34a) on the nanoneedles (green); scale bars, 10 μm for three-dimensional and top view, 1 μm for enlarged view. (I) Comparison of miRNA (let-7a or miR-34a) signals from different controls. (J) Ratio of signal+ nanoneedles for experiments with or without HCR amplification. For (D), (E), (I), and (J), n = 3; the error bar indicates SEM; *P < 0.001 by ANOVA test.

Temporal miRNA transcriptome

The capability to capture the dynamics of miRNA expression is extremely important for a profiling technique, as intracellular miRNAs play a key regulatory role in gene expression networks and change when cells switch their status (27). We then applied inCell-Biopsy to characterize relevant miRNAs in cells undergoing DNA damage or at different stages of a cell cycle to show its potential as a technique for probing cellular dynamics. Upon ultraviolet-induced DNA damage, let-7a was significantly down-regulated, and miR-16 and miR-26 were significantly up-regulated within just several minutes (fig. S3). In an extended temporal window as the cells progress to different division cycles, let-7a, miR-21, and miR-34a were observed to gradually increase from G1, to S, to G2 stage, while miR-24 remained stable at these stages (fig. S4). These results echo well with the literature (28, 29) and demonstrated inCell-Biopsy’s capability to monitor the fluctuation of miRNAs expression associated with cellular activity. In addition, the technique not only provides an overall assessment of miRNA expression in the cells but also captures the dynamic heterogeneity of cell populations, as confirmed by flow cytometry (fig. S4D) and qRT-PCR analysis (fig. S4, F and G).

Heterogeneity evolution over mouse ESC differentiation

After the above technical validations, we next applied inCell-Biopsy to investigate the temporal miRNA transcriptome and its relationship to cellular heterogeneity over the differentiation of mouse ESCs (mESCs) (HB9: GFP) toward motor neurons. We chose to profile nine different miRNAs in these cells at day 0, day 7, and day 14 from the induction of differentiation (Fig. 3, A and B). By using a custom-developed processing streamline and program (fig. S5), the expression of the nine miRNAs at various differentiation stages were obtained (Fig. 3C). When all the data were pooled together blindly, t-distributed stochastic neighbor embedding (t-SNE) (30) quantification showed the overall evolutionary change of the cells by the appearance of three self-organized clusters along with ESC differentiation, suggesting the validity of using combinatory miRNA expression pattern to indicate cell identity in this process (Fig. 3D).

Fig. 3 Temporal profiling of miRNA by inCell-biopsy.

(A) Experimental design of monitoring miRNA dynamics over the differentiation of embryonic stem cells (ESCs). RA, retinoic acid; and SAG, smoothened agonist. (B) Phase (Ph)–contrast and fluorescence (green fluorescence indicates GFP) images showing morphological change of the cells along with differentiation. Scale bars, 50 μm. (C) Confocal fluorescence image (top view) of diamond nanoneedles after fishing and HCR amplification. Scale bar, 20 μm. The boxed region is enlarged below to show the expression of nine miRNAs from three rounds of amplification and visualization. Scale bars, 2 μm. (D) t-SNE clustering of the pooled multidimensional miRNA vectors that resulted from inCell-biopsy at all three stages, showing the overall evolution of miRNA expression along with ESC differentiation.

Cells generated from ESC differentiation are typically heterogeneous (31), the inCell-Biopsy technique provides us the possibility to decipher this heterogeneity and biogenic evolution by using temporal miRNA dynamics (16). For each of the nine miRNA targets, we quantified its fold change at later differentiation stages with respect to the initial stem cell level, performed a self-diffusion–based spectral clustering (details in Materials and Methods Section) for the multidimensional miRNA measurements from thousands of nanoneedles pooled from six independent replicates, and determined the optimal numbers of clusters by eigengap (fig. S6A) (32). It was found that stable subpopulations were clearly observed at both 7 and 14 days after differentiation, and the cells appeared to be more scattered at the later stage (Fig. 4, A and B). The clustering results were also confirmed by t-SNE analysis (Fig. 4C) as well as principle component analysis (fig. S6B). The heatmaps and violin plots of the nine miRNAs showed the unique expression pattern of each cluster at a specific differentiation stage (Fig. 4, D, E, G, and H and fig. S7) and also suggested some similarities of particular clusters across different stages (e.g., cluster 3 at day 7 versus cluster 5 at day 14), implicating potential evolutionary correlation between them. Statistically, we were able to identify cluster-specific miRNA expression signatures (table S2). The shared signature between clusters of day 7 and day 14 supports our speculation of their evolutionary relationship. For instance, cluster 3 of day 7 and cluster 5 of day 14 shared the similar signature miRNAs of miR99a, miR218, and miR9. miR24, miR218, and miR219 are the shared signature miRNAs of cluster 4 of day 7 and cluster 3 of day 14. The discovery of these clusters was made by statistical analysis of signals from thousands of nanoneedles that were interfaced with a large population of cells. The proportion of a particular cluster (out of all nanoneedles) was also evaluated as an indicator for the percentage of a cell subpopulation, assuming an even distribution of the nanoneedles on a uniform culture of cells (Fig. 4F). This scattering information would be lost if averaged miRNA measurements were performed with cell lysate as what are mostly done by existing methods.

Fig. 4 The dynamic evolution of mESC heterogeneity revealed by inCell-biopsy.

Self-diffusion–based spectral clustering and associated similarity network for the multidimensional miRNA measurements from thousands of nanoneedles at day 7 (A) or day 14 (B) of differentiation. (C) Separation of cellular subpopulations indicated by t-SNE analysis at day 7 or day 14. Heatmap showing distinct miRNA expression patterns between different clusters obtained from unsupervised classification at day 7 (D) or day 14 (E). (F) Sector graph showing the proportion of nanoneedles in each cluster out of the total number of nanoneedles on day 7 or day 14. Radar plots (left) and associated violin plots (right) show the averaged expression of the nine miRNAs for each of the identified clusters at day 7 (G) or day 14 (H).

Mapping of miRNA transcriptome to cellular composition

To study the evolutionary correlation among different cell subpopulations (represented by the clusters) over differentiation, we used the multidimensional miRNA data from inCell-Biopsy to study the statistical association between the clusters of day 7 and day 14 to determine the closest pairs. Each cluster of day 14 can be uniquely traced back to link with a cluster of day 7 (P < 0.001, hypergeometric tests; Fig. 5A). For example, cluster 3/4/5 of day 14 was respectively found to be most correlated with cluster 4/2/3 of day 7; these paired clusters also showed similar miRNA expression patterns as shown in the violin plots (Fig. 4, G and H). Clusters 1 and 2 of day 14 both traced back to cluster 1 of day 7, suggesting cluster 2 of day 14 to be a newly differentiated subpopulation derived from cluster 1 of day 7 (Fig. 5B).

Fig. 5 Mapping of miRNA transcriptome to cellular composition.

(A) Hypergeometric tests for determining the closest pair of clusters from the two differentiation stages. Significant associations are labeled by red squares, colored in proportion to −log10(P value) (all P < 0.001). (B) The phylogenetic tree shows the evolutionary relationship among the clusters (cell subpopulations) as differentiation proceeds. The widths of the branches are proportionate to transformed P values [−log10(P values)] derived from hypergeometric tests. (C) Correlation between the cluster miRNA pattern (derived from inCell-biopsy) with the results acquired by miR-seq for sorted motor neurons/progenitors. *P < 0.001. (D) Quantitative analysis of the averaged expression of the nine miRNAs for motor neuron–like clusters at day 7 and day 14. Error bars indicate SEM from six independent experiments. (E) Density histograms and associated violin plots showing the variation and distribution of nine miRNAs expression for motor neuron–like clusters at day 7 and day 14.

To figure out the relationship between the identified cluster based on miRNA expression and the cell type identity, we then focused on differentiated motor neuron/progenitors (GFP+) and compared the inCell-Biopsy–acquired miRNA profile for each cluster to the data acquired by miRNA sequencing (miR-seq) for sorted motor neuron/progenitors at various differentiation stages. We found that the majority of the nine-dimensional miRNA vectors in cluster 3 of day 7 or cluster 5 of day 14 showed significantly higher correlation to the miR-seq–measured miRNA expression pattern (P < 0.001, Wilcoxon signed-rank test; Fig. 5C), which was not observed for the other clusters of the same differentiation stage, suggesting that the nanoneedles in cluster 3 of day 7 or cluster 5 of day 14 mostly sampled motor neurons/progenitors in the miRNA biopsy operation. Taking a closer look at the two clusters, we further observed dynamic changes of different miRNAs (Fig. 5, D and E). For example, miR-294, a stem cell–specific miRNA (33), was significantly reduced from day 7 to day 14, whereas the motor neuron–enriching miR-9 and miR-218 (34) were significantly increased over the same period (Fig. 5D). In addition, for the two motor neuron-like clusters, the miRNA expression was generally more scattered at day 14 (compared with day 7; Fig. 5E), which suggested an increased variation of the cell status as differentiation progresses.


Here, we develop a highly versatile and powerful technique, inCell-Biopsy, for in situ–multiplexed profiling of miRNAs in living cells. The technique is capable of cherry picking targeted miRNAs from cell cytoplasm while leaving the samples intact afterward. For quantitative analysis of cellular RNAs, most of the existing techniques (e.g., qRT-PCR, microarray, RNA sequencing) started with RNA samples extracted from a population of cells and only provide an averaged measurement of the cell population (13, 14). InCell-Biopsy, on the other hand, isolates targeted miRNAs from a large number of individual cells within just a few minutes by using a diamond nanoneedle–facilitated molecular fishing system (19), and parallels in situ amplification, visualization, and quantification of miRNAs using each nanoneedle as a separated reaction chamber. In this way, our method not only detects averaged miRNA expression level but also captures the cellular heterogeneity of a cell population based on miRNA profiling, which is typically missed in other methods and only accessible using single-cell RNA sequencing (scRNA-seq) analysis (16). In this study, the density of the diamond nanoneedles was roughly controlled at ~5 nanoneedles per 10 by 10 μm2 region. Although we cannot establish an exact one-to-one (or multiple-to-one) contacting map between the nanoneedles and each individual cell, the inCell-Biopsy technique enables a quasi–single-cell analysis to provide rich information for characterizing cell mixtures by using multidimensional miRNA profiles. As a proof of concept, we used the inCell-Biopsy technique to dissect cellular heterogeneity over the differentiation of ESCs and investigated the evolution of the cells with a dynamic temporal miRNA transcriptome analysis.

While the intracellular biopsy strategy has been previously reported (19, 3537), in this study, such a concept was further elaborated with specific biochemical design targeting multiple miRNAs, along with a complete framework for multiplexed in situ signal amplification, visualization, and quantification, which altogether are formulated as a quasi–single-cell miRNA profiling platform. Notably, the diamond nanoneedles are rigid enough to puncture cell membrane and remain ultraelastic at nanoscale to sustain the deformation without fracture during an inCell-biopsy operation (38). Although different nanostructures were recently developed as tools to isolate intracellular materials from living cells (3537) and may have the potential to be used for detecting miRNAs when combined with sequencing techniques, our inCell-biopsy technique stands out with a balanced combination of in situ capability, high throughput, ease of use, and independence of expensive equipment.

As one of the core merits, the inCell-Biopsy does not involve any cell lysis and RNA preparation procedures; therefore, the examined cells can be preserved for further longitudinal analysis. This feature also markedly simplifies the experimental operation, reduces the processing time, and provides the opportunity to quantitatively examine the temporal dynamics of miRNA expression for the same batch of cells receiving external stimuli or undergoing internal switch of cellular programs. It would be extremely useful when miRNA profiling is used as a characterization or quality control for cell-based therapeutic treatment (39). Meanwhile, the capability to directly fish miRNAs from the cytoplasm of an individual cell effectively bypasses the dilution of low-abundance miRNAs and prevents sample loss during cell lysing and RNA extraction procedures. Although only a single copy of miRNA is presented in a cell, the actual concentration for nanoneedle-assisted inCell-Biopsy would be around 10−13 to 10−12 M, which is well tolerated by the detection limit of the technique (10−15 M).

Our inCell-Biopsy is based on an intracellular molecular fishing system, in which diamond nanoneedles are used as the fishing rod and RNA binding protein (p19) is used as the fishing hook, which specifically binds to dsRNA (not to ssRNA or dsDNA) in a size-dependent manner (20). For an inCell-Biopsy operation, the dsRNA complexes are formed by the hybridization between the targeted single-strand miRNA and a complementary bait sequence, which was thought to diffuse into cell cytosol via the nanopuncture-induced reversible membrane disruption (23). It is also possible that cytosolic materials could diffuse to cells outside, but this should not be an issue for the treated cells in this study, as the nanoneedles were tightly interfaced with the mobile lipid bilayer membrane (fig. S1), making it less likely for intracellular components (e.g., miRNAs) to leak out and be captured. Notably, intracellular pri-miRNAs or pre-miRNAs that lead to false positive in traditional PCR-based detections (13) would not interfere with our assay, because their structures would prohibit bait hybridization and subsequent binding to p19 proteins. The introduction of an encoded bait sequence for each miRNA target further enhances the specificity of the inCell-Biopsy technique. Practically, p19 can bind to all available dsRNAs of the right length, so that multiplexed detection (e.g., nine miRNAs in this study) can be easily implemented by effortless intracellular delivery of multiple bait sequences (20). Compared with scRNA-seq, the throughput of the inCell-Biopsy technique may be lower at the current stage, but it has, undoubtedly, advantages in substantially lower cost and more efficient experimental protocol. For improvement, the fluorescence labeling system can also be fine-tuned to include more channels to improve the assay throughput. For example, if the hairpin sequences were labeled with quantum dots, it would be easy to achieve an eight-channel imaging system, and three rounds of imaging would increase the throughput to 21 miRNA targets. In addition, the incorporation of certain barcoding strategy (e.g., nanostring system) (40) can further increase the analytical throughput to allow the analysis of hundreds of miRNA targets within a single visualization cycle.

As we had demonstrated, a nine-dimensional miRNA vector space produced by inCell-Biopsy already carries rich information for the identification of heterogeneous clusters that represent the cellular subpopulations differentiated from ESCs. The clustering was autonomously derived from a quasi–single-cell analysis of the miRNA expression patterns, which have recently been reported to be a good indicator for cellular heterogeneity (11, 12). While we cannot spatially determine the contacting relationship of each nanoneedle to an individual cell, statistically, the miRNA profiling by inCell-Biopsy can still reflect the compositional nature of examined cells, assuming a random but uniform distribution of the diamond nanoneedles over an evenly cultured cell. The multidimensional miRNA vector derived from each nanoneedle is treated as an input to a huge miRNA vector space, which effectively created a quasi–single-cell analysis framework for miRNA transcriptome analysis. Particularly, in this study, the correspondence between nanoneedle clusters and identity of specific cell subpopulation was verified to further confirm the validity of the analytical results acquired by inCell-Biopsy. For example, we used a protocol to direct motor neuron differentiation and blindly identified a major nanoneedle cluster (out of all nanoneedles) that was highly correlated to motor neuron identity at both day 7 and day after differentiation. However, when the same mESCs (HB9: GFP) were undergoing spontaneous differentiation without the induction compounds (retinoic acid and smoothened agonist), the motor neuron-like cluster was not discoverable (fig. S8).

The capability to retain cell sample after inCell-Biopsy operation enables multiround miRNA profiling at different time points, thus providing a temporal miRNA transcriptome analysis. Our results show that the inCell-Biopsy not only creates a quick snapshot of the heterogeneity of the examined live cells based on their miRNA profiles but also captures the temporal dynamics of miRNA expression, and it subsequently gives the cellular evolutionary path, as well as the biogenic relationship among heterogeneous emerging cell populations, which is especially informative for clinical applications (7).

In summary, we demonstrate a novel and powerful technique, inCell-Biopsy, for profiling miRNAs in living cells. The temporal miRNA dynamics captured by this technique can be used to reveal the evolution of cellular heterogeneity in mixed cell populations over extended culture periods, potentially providing a quick and convenient evaluation platform for the quality control of the emerging therapeutic strategies involving cell components.


Cell culture

HB9: GFP mESCs were acquired from the Stem Cell Core Facility of Columbia University. ESCs were seeded in a petri dish coated with 0.1% gelatin and were further cultured in an incubator at 37°C with 5% CO2 for proliferation. After 3 days, ESCs were trypsinized for cell seeding. Typically, 250 ml of ESC culture medium consisted of 200 ml of EmbryoMax Dulbecco's modified Eagle's medium (DMEM) (Millipore), 37.5 ml of fetal bovine serum (FBS; Hyclone), 2.5 ml of EmbryoMax MEM Nonessential Amino Acids (Millipore), 2.5 ml of nucleosides (Millipore), 2.5 ml of 200 mM l-glutamine (Invitrogen), 2.5 ml of penicillin/streptomycin (pen/strep) (10,000 U/ml penicillin and 10,000 μg/ml streptomycin, Invitrogen), 180 μl of diluted 2-mercaptoethanol [diluted 1:100 in phosphate-buffered saline (PBS) with Mg and Ca; Invitrogen], and 25 μl of leukemia inhibitory factor (LIF)/ESGRO (Millipore). Afterward, the embryonic stem medium was replaced by differentiation medium. Typically, 450 ml of differentiation medium consisted of 200 ml of Advanced DMEM/F12 (Invitrogen), 200 ml of Neurobasal Medium (Invitrogen), 46 ml of KnockOut Serum Replacement (Invitrogen), 4.6 ml of pen/strep, 4.6 ml of l-glutamine, and 320 μl of diluted 2-mercaptoethanol. After 2 days, retinoic acid (RA; diluted 1:1000 in differentiation medium; Sigma-Aldrich) and smoothened agonist (SAG; dilutes 1:1000 in differentiation medium; Sigma-Aldrich) were added into the medium for a strong induction to motor neuron differentiation. After 3 days of in vitro differentiation, the differentiation medium supplemented with 4.5 μl of glial-derived neurotrophic factor (Invitrogen), 9 ml of B27 (Invitrogen), and 4.5 ml of N2 supplement (Invitrogen) was used for a better motor neuron growth.

A549 cancer cells were maintained in DMEM (Life Technologies) supplemented with 10% FBS, l-glutamine, and pen/strep. Before molecular fishing experiments, cells were seeded in a four-well multidish (Nunclon) and allowed to grow until ~80% confluence.

Fabrication and characterization of diamond nanoneedles

The fabrication of diamond nanoneedles follows a protocol as previously described (19), involving the deposition of nanodiamond film and subsequent bias-assisted reactive ion etching (RIE) by electron cyclotron resonance microwave plasma chemical vapor deposition (ECR-MPCVD). N-type (001) silicon wafers of 3-inch diameter were used as a substrate. Before nanodiamond deposition, the substrate was soaked in ultrasonic bath for 60 min in ethanol, containing a suspension of nanodiamond powders with a grain size of 5 nm. Nanodiamond films 7 μm thick were then deposited using a commercial ASTeX MPCVD equipped with a 1.5-kW microwave generator. The nanodiamond deposition was performed in the plasma induced in 10% CH4/H2 mixture at a total pressure of 30 torr and total gas flow rate of 200 standard cubic centimeter per minute (SCCM). The microwave power and deposition temperature were maintained at 1200 W and 800°C, respectively. After the nanodiamond film deposition, RIE was performed using ECR-MPCVD. The ASTeX microwave source used a magnetic field of 875 G generated by an external magnetic coil. The RIE used a mixture of 45% Ar and 55% H2 as the reactive gases, which were supplied at a total flow rate of 20 SCCM. The substrate bias was −200 V, and the reactant pressure was 7 × 10−3 torr. The etching duration was 3 hours, and the input microwave power was 800 W. The morphology of the resulted diamond nanoneedles was characterized by scanning electron microscopy (SEM; Philips FEG SEM XL30), and the sample was tilted 90° for SEM.

Functionalization of diamond nanoneedles

To functionalize the diamond nanoneedles with p19 protein, a patch was first bathed in piranha (3:1, v/v; 98% H2SO4:27.5% H2O2) solution at 90°C for 1.5 hours and then cleaned by distilled water, methanol, methanol/dichloride methane (DCM) mixture (3:1, v/v), and DCM sequentially. The nanoneedle patch was dried with nitrogen and then immersed in (3-aminopropyl)triethoxysilane solution (20% in DCM, v/v) overnight in a nitrogen-protected environment. Ethanol, isopropyl alcohol, and distilled water were sequentially used to rinse the nanoneedle patch, which would be further dried by nitrogen blow. The nanoneedle patch was then bathed in NHS-biotin solution (1 μg/ml in PBS; Sigma-Aldrich) for 1 hour, streptavidin (20% of the streptavidin was labeled by fluorescent dye, Alexa Fluor 488) solution (10 μg/ml in PBS; Invitrogen) for 2 hours, and biotinylated p19 siRNA binding protein solution (1 μg/ml in PBS; New England Biolabs) for 1 hour. The nanoneedle patch was rinsed with distilled water between adjacent bath steps. After each experiment, the nanoneedle patch was soaked in hot (~90°C) piranha solution (3:1, v/v; 98% H2SO4:27.5% H2O2) to remove all cross-linked materials (protein, nucleotides, etc.) on the nanoneedle surface, and SEM images were taken ensure the integrity of the nanoneedle structure. In this way, one patch can be used for at least 20 times. All the materials were acquired from Sigma-Aldrich unless otherwise specified.

Centrifugation-controlled molecular fishing

Intracellular delivery of RNA bait sequences and miRNA fishing were performed using a centrifugation-controlled process (19). For cells in adherent culture, the medium was first removed, and 100 μl of RNA bait sequence solution (10 nM for every bait sequence in serum-free medium) was applied to the cells. A nanoneedle patch was then placed facing toward the cells in a four-well petri dish. The whole complex was then placed in a centrifuge with a plate rotor and spun at 400 revolutions per minute (rpm) (22.8 g) for 4 min. After first centrifugation, the setup was placed in an incubator for 10 min to allow the bait sequences to diffuse into cytoplasm and to form dsRNA with their intracellular targets. Afterward, a second centrifugation was performed to enhance miRNA fishing results. During a centrifugation, the ramping speed was controlled to maintain a smooth acceleration to avoid any movement of the nanoneedle patch on cells. For acceleration and deceleration, 3 and 6 rpm/s were, respectively, selected.

In situ HCR amplification

Each RNA bait sequence used in the inCell-Biopsy has a unique 10-bp overhang sequence that can be used to amplify the signal coupling with the HCR (41). To perform the on-needle HCR amplification of miRNAs, the initiator sequences were diluted in the hybridization buffer containing 5× sodium citrate (SSC; Invitrogen) with 0.05% tween (pH 7.4; Invitrogen) to a final concentration of 10 nM; the hairpins were diluted in reaction buffer to a final work concentration of 20 nM. After a rinse with 0.05% SDS for 15 min, the nanoneedle patch was immersed in the initiator solution for hybridization between the dsRNAs and the initiator sequences. Following a quick rinse with the wash buffer (1× SSC, 0.05% tween), the patch was further incubated at 37°C for 3 hours in hybridization buffer containing different hairpin DNAs (20 nM) with FAM, JOE, CY3, or CY5 fluorescent labels and black hole quencher. After the initiation of the HCR, the heterodimer was separated, and the absorption/emission of the fluorophore was restored. To guarantee the separation and reduce the requenching effect after the binding of hairpins 1 to 2, a short unmatched sequence was included in the hairpin sequences to work as a spacer. All hairpin and initiator oligonucleotides were acquired from BGI (Shenzhen, China) and summarized in table S1.

DNase I–assisted multiround signal amplification

After first imaging, the HCR-amplified nanoneedle patch was immersed into DNase I solution for 1 hour in 37°C to fully elute the DNA hybridized on the needle surface, followed by washing three times with wash buffer. After that, the nanoneedle patch was enabled to perform the HCR amplification to detect another four miRNA targets.

Confocal microscopy and image processing

After each HCR amplification process, confocal microscopy (Leica SP8, 40× objective with 1.3 numerical aperture, water immersion) was performed to visualize and quantify the miRNAs captured on the surface of diamond nanoneedles. A nanoneedle patch was scanned with 0.3-μm z resolution to get a stack of 45 to 55 slices, and both three-dimensional reconstruction and maximal projection of the stack were acquired. As a result of the HCR amplification, the fluorescent speckles (from DNA hairpin sequences) on nanoneedle surface was quantified and used to analyze the fishing of miRNA targets from the cells inside. After three rounds of amplification, images were projected and aligned to get a 12-channel image stack containing the fluorescence intensity and relative position information. To differentiate the positive signals from background noises or debris, fluorescent speckles of 0.4 to 1.5 μm in diameter were firstly selected as nanoneedle regions, and a fluorescence threshold was then applied to sort out the positive nanoneedles with captured miRNAs. For each nanoneedle patch, we lastly obtained a matrix of intensities representing miRNA expression levels, where the columns are miRNA targets and the rows are different nanoneedles.

Data preprocessing and clustering

After obtaining the miRNA expression matrix, we divided the expression data of day 7 and day 14 by matched average value of that at embryo stem cell stage followed by log2 transformation to derive miRNA expression fold change data. To make sure that the miRNA expression data are comparable between different replicates, quantile normalization was subsequently performed for day 7 and day 14, respectively.

Having performed the preprocessing, the miRNA expression fold change data from all the replicates were combined as input for unsupervised classification at day 7 and day 14, respectively. We performed self-diffusion analysis (42), which was implemented by propagating the affinity matrices to improve sample similarities learning, followed by spectral clustering (32) for its relatively better adaptability to data distribution and the lower time consumption. We calculated eigengap (32) based on local scaling affinity, which infers the self-tuning affinity of sample-by-sample distances, and chose the optimal number (k) of clusters for clustering (fig. S6A). The sample-by-sample similarity matrices and similarity networks for day 7 and day 14 are shown in Fig. 4 (A and B), respectively.

Evolutionary relationship and determination of cell components

For each miRNA profiling at day 7 (or day 14), we calculated Pearson correlation coefficients (PCCs) between its miRNA expression and the average expression levels of nanoneedles in different clusters at day 14 (or day 7). The nanoneedle at day 7 was subsequently assigned to the most correlated cluster at day 14 and vice versa. A confusion matrix was constructed to summarize the total number of nanoneedles classified simultaneously to each pair of clusters at day 7 and day 14, and a hypergeometric test was subsequently performed to evaluate the statistical significance of their association. First, we calculated the average expression profile of probes in each cluster of day 7. Second, for each nanoneedle in day 14, we calculated PCC with the average profile of each cluster of day 7, and the nanoneedle (day 14) was paired to the cluster (day 7) with the highest PCC. Third, all paired day 14–day 7 nanoneedle relationships were counted, summarized, and followed by a hypergeometric test for overrepresentation of nanoneedles in day 14 paired to a cluster in day 7. Last, P values derived from the hypergeometric tests were adjusted for multiple hypothesis testing using the Benjamini-Hochberg procedure and illustrated as a heatmap. Similarly, using clusters of day 14 as a reference, we did association tests for the nanoneedles of day 7, and the conclusion is highly consistent.

To illustrate the potential evolutionary relationship between clusters at different stages, a phylogenetic tree was generated on the basis of the statistical associations between clusters at day 7 and day 14, where the branch width represents the transformed P value [−log10(P)] derived from the hypergeometric tests. For validation, motor neuron cells (GFP+) were isolated with Sony SH800S cell sorter; the total RNA of sorted cells was extracted using the TRIzol reagent kit (Life Technologies) for miR-seq analysis (BGI). To investigate the potential cell type identities of nanoneedle clusters, we calculated PCCs between the inCell-Biopsy–acquired miRNA profiles and the expression levels of the nine miRNAs measured by miR-seq.

qRT-PCR of miRNAs

To measure miRNAs (let-7a, miR-21, miR-24, miR-34a, and RNU43) using qRT-PCR, the total RNA was extracted from A549 cells using the TRIzol reagent kit (Life Technologies). For 15-μl reactions, 10 ng of total RNA was reverse-transcribed and analyzed by the TaqMan miRNA Assays kit (product no. 4366596; Life Technologies). The expression of a particular miRNA was analyzed using the Applied Biosystems real-time PCR instrument following the manufacturer’s protocol.

Statistical analysis

At least three independent biological replicates were used for all experiments (n ≥ 3); for each replicate, signals from at least 250 nanoneedles were collected for analysis. For Fig. 2C, the bin size along the y axis was 60 (fluorescence, arbitrary units) for the violin plots; the whisker range of the overlaying boxplots is 1 to 99%, and each box shows 25, 50, and 75% percentile of the data. Kruskal-Wallis analysis was performed to determine the statistical significance among different dsRNA concentrations, and P < 0.005 indicates a significant difference. For Fig. 2 (H and I), analysis of variance (ANOVA) was performed to determine the statistical significance; P < 0.05 indicates a significant difference. The error bars indicate SEM from three independent experiments. For Fig. 4 (G and H), the bin size along the y axis was 0.1 (fold change of the expression level) for the violin plots. To identify cluster-specific miRNA expression signatures, a two-tailed Student’s t test was performed to assess whether each miRNA is differentially expressed between a specific cluster and the other clusters of day 7 (or day 14). For each cluster of day 7 (or day 14), the miRNA expression signatures were prioritized on the basis of the absolute log2 expression level (|log2EL| > = 0.75) and Benjamini-Hochberg–adjusted P value (P < 0.001). For Fig. 5C, Wilcoxon signed-rank test was performed to determine the statistical significance; P < 0.001 indicates a significant difference. For Fig. 5D, the error bars indicate SEM from six independent experiments. For Fig. 5E, the bin size along the y axis was 0.1 (fold change of the expression level) for the violin plots.


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: Funding: This work was supported by the National Natural Science Foundation of China (81871452, 81802384, and 51772318), by the Science, Technology and Innovation Committee of Shenzhen Municipality (JCYJ20170818100342392, JCYJ20180507181624871, and JCYJ20170413141236903), by the General Research Fund (11278616, 11203017, 11102317, 11103718, and 11103619) from the Research Grants Council of Hong Kong SAR, and by the Health and Medical Research Fund (06172336) from the Food and Health Bureau of Hong Kong SAR. Author contributions: P.S. conceived the project, designed, and supervised the research. Z.W., X.Z., and K.X. carried out the experiments and analyzed the data. L.Q. and X.W. performed the statistical and bioinformatics analysis. Y.Y. and W.Z. provided the diamond nanoneedle array. M.L., E.H.C.C., X.J., and L.H. helped with the experiments. All authors contributed to the writing of the manuscript. Competing interests: P.S., W.Z., L.H., X.W., and Z.W. are inventors on a pending patent related to this work (no. US15/875,385, filed 18 January 2018). The authors declare that they have no competing interests. Data materials and availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.
View Abstract

Stay Connected to Science Advances

Navigate This Article